WEBVTT - Adobe Chair & CEO Shantanu Narayen Talks AI Video Generator

0:00:02.440 --> 0:00:06.800
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news.

0:00:07.320 --> 0:00:12.039
<v Speaker 2>Adobe makes its AI video debut, announcing new tools allowing

0:00:12.080 --> 0:00:14.960
<v Speaker 2>for video generation now. The company calls it Firefly, and

0:00:15.000 --> 0:00:17.639
<v Speaker 2>it's set to be implemented in its video editing software

0:00:17.640 --> 0:00:20.320
<v Speaker 2>Premiere Pro, taking on, of course, the likes of OpenAI,

0:00:20.440 --> 0:00:24.400
<v Speaker 2>Meta Google a place to say. Adobe CEO Shantanu Orian

0:00:24.560 --> 0:00:28.800
<v Speaker 2>joins us now on the back of the announcement in Miami. Shantanu,

0:00:29.000 --> 0:00:32.080
<v Speaker 2>what's interesting is you're getting into the hands of users

0:00:32.280 --> 0:00:36.440
<v Speaker 2>faster than rivals. Is it based on what it's trained upon?

0:00:36.600 --> 0:00:38.440
<v Speaker 2>How are you able to give it to us so quick?

0:00:40.280 --> 0:00:42.800
<v Speaker 1>Caroline, first, thanks for having me on the show.

0:00:42.920 --> 0:00:43.240
<v Speaker 3>Max.

0:00:43.280 --> 0:00:47.000
<v Speaker 1>As you know, is our biggest creativity event, and it's

0:00:47.040 --> 0:00:50.240
<v Speaker 1>the time that our engineers really take it upon themselves

0:00:50.880 --> 0:00:54.880
<v Speaker 1>to show their magic to the entire creative community. As

0:00:54.920 --> 0:00:58.160
<v Speaker 1>you point out, we have been hard at work ensuring

0:00:58.200 --> 0:01:01.520
<v Speaker 1>that our Firefly set of models works on things like

0:01:01.600 --> 0:01:05.040
<v Speaker 1>images and vector and design. But today's a big day

0:01:05.160 --> 0:01:10.280
<v Speaker 1>because we have been releasing our Firefly video models. You know,

0:01:10.319 --> 0:01:12.880
<v Speaker 1>our strategy has always been that it's not just about

0:01:12.880 --> 0:01:16.080
<v Speaker 1>the models that people develop, but it's the integration with

0:01:16.200 --> 0:01:19.200
<v Speaker 1>the apps where the real magic is, and so we've

0:01:19.200 --> 0:01:23.000
<v Speaker 1>had thirteen billion generations where people have used either Photoshop

0:01:23.080 --> 0:01:25.920
<v Speaker 1>or illustrator to do it, and today that magic comes

0:01:25.920 --> 0:01:29.360
<v Speaker 1>to video in Premiere Pro, will be in after effects

0:01:29.400 --> 0:01:33.040
<v Speaker 1>as well as on firefly dot com Chantanu.

0:01:33.840 --> 0:01:37.000
<v Speaker 3>The most interesting part is the accessibility. There are no

0:01:37.240 --> 0:01:41.479
<v Speaker 3>usage limits one, but beyond the basic subscription fees, you're

0:01:41.520 --> 0:01:46.000
<v Speaker 3>not charging for the AI tools. Will that change.

0:01:47.000 --> 0:01:50.040
<v Speaker 1>Well, Our focus really first is on ensuring that we

0:01:50.240 --> 0:01:53.600
<v Speaker 1>developed these great models. We're very differentiated, and these models

0:01:53.600 --> 0:01:56.920
<v Speaker 1>were designed to be commercially safe, so they've only been

0:01:56.960 --> 0:02:01.400
<v Speaker 1>trained on licensed and public domain content, and you know,

0:02:01.440 --> 0:02:03.520
<v Speaker 1>we want people to experiment with it. So I think

0:02:03.600 --> 0:02:07.840
<v Speaker 1>clearly on the imaging, vector and design front, we have

0:02:07.920 --> 0:02:09.960
<v Speaker 1>said as part of the subscription, we don't want you

0:02:10.000 --> 0:02:13.000
<v Speaker 1>to think about credits. We just want you to experience

0:02:13.040 --> 0:02:15.600
<v Speaker 1>the magic that exists. A video might be a little

0:02:15.639 --> 0:02:18.040
<v Speaker 1>bit different ed. What we've announced today is that it's

0:02:18.080 --> 0:02:21.720
<v Speaker 1>available in a beta form. The cost of producing video

0:02:21.880 --> 0:02:25.400
<v Speaker 1>is clearly a little bit more expensive, but first steps

0:02:25.440 --> 0:02:27.240
<v Speaker 1>is to just get it in the hands of people,

0:02:27.600 --> 0:02:31.240
<v Speaker 1>have them try it, give us feedback, make sure it's specific,

0:02:31.280 --> 0:02:33.760
<v Speaker 1>and then as I think the amount of video that's

0:02:33.840 --> 0:02:38.760
<v Speaker 1>being generated is expanding. We will probably have different pricing

0:02:38.840 --> 0:02:39.560
<v Speaker 1>models for that.

0:02:41.200 --> 0:02:43.720
<v Speaker 3>One of the features is to take an existing piece

0:02:43.720 --> 0:02:47.800
<v Speaker 3>of video and use generative AI to extend it, extend

0:02:47.840 --> 0:02:51.359
<v Speaker 3>beyond actual. As someone that trained in their early career

0:02:51.400 --> 0:02:55.560
<v Speaker 3>on premiere, I'm excited to use that. How technically difficult

0:02:55.680 --> 0:02:58.880
<v Speaker 3>was it for Adobe to achieve that versus the work

0:02:58.880 --> 0:02:59.760
<v Speaker 3>you've done in photo?

0:03:01.520 --> 0:03:04.639
<v Speaker 1>Well, every time you show something in photo and people say,

0:03:04.680 --> 0:03:08.920
<v Speaker 1>oh my god, that's magic. To your point, the ability

0:03:08.960 --> 0:03:12.480
<v Speaker 1>to extend that into video across every frame is so

0:03:12.600 --> 0:03:16.640
<v Speaker 1>much more computationally intensive as well as challenging in order

0:03:16.680 --> 0:03:19.399
<v Speaker 1>to do. But the use cases are so dramatic. You've

0:03:19.400 --> 0:03:22.880
<v Speaker 1>had a video for some reason, unfortunately it got clipped

0:03:22.880 --> 0:03:25.440
<v Speaker 1>off and you just want to extend it. You have

0:03:25.560 --> 0:03:28.919
<v Speaker 1>video that you want to align with audio, the song

0:03:29.040 --> 0:03:30.960
<v Speaker 1>is not long enough. How do you do it? So?

0:03:31.040 --> 0:03:35.920
<v Speaker 1>I think this generative extent concept, which is really something

0:03:35.960 --> 0:03:38.960
<v Speaker 1>that's very pervasive in video editing, is something that the

0:03:39.000 --> 0:03:41.600
<v Speaker 1>customers will really like. The Other thing that we've done

0:03:42.040 --> 0:03:45.040
<v Speaker 1>is because we can now analyze every frame. When you

0:03:45.080 --> 0:03:48.960
<v Speaker 1>want to remove something that inadvertently came into the image,

0:03:49.080 --> 0:03:51.720
<v Speaker 1>you don't have to go through the entire process again,

0:03:52.000 --> 0:03:54.600
<v Speaker 1>you can just through post production change it. So some

0:03:54.640 --> 0:03:58.560
<v Speaker 1>pretty incredible stuff and we're really excited to see what

0:03:58.600 --> 0:03:59.720
<v Speaker 1>our customers do with this.

0:04:00.280 --> 0:04:06.040
<v Speaker 2>You talked about the computational feat the intensity, What about GPUs,

0:04:06.160 --> 0:04:08.920
<v Speaker 2>what about your own access to compute?

0:04:10.200 --> 0:04:13.440
<v Speaker 1>You know, well, we've partnered with you know, Nvideo for

0:04:13.800 --> 0:04:17.080
<v Speaker 1>years in making sure all of this stuff is available.

0:04:17.720 --> 0:04:19.960
<v Speaker 1>You know, it was great to hear from Jensen this

0:04:20.080 --> 0:04:24.000
<v Speaker 1>morning when he heard about our new announcements. But you know,

0:04:24.040 --> 0:04:27.240
<v Speaker 1>we're partnering with all of the chip manufacturers as well

0:04:27.279 --> 0:04:30.600
<v Speaker 1>as all of the hyper scalers because our intent is

0:04:31.000 --> 0:04:34.560
<v Speaker 1>not just to make this magic available through you know,

0:04:34.600 --> 0:04:38.440
<v Speaker 1>those chips, but also increasingly people are talking about how

0:04:38.480 --> 0:04:41.039
<v Speaker 1>you can make this available on a mobile device or

0:04:41.080 --> 0:04:44.080
<v Speaker 1>in a hybrid environment. So you know, as long as

0:04:44.120 --> 0:04:47.640
<v Speaker 1>they keep performing in terms of what they are doing

0:04:47.880 --> 0:04:51.520
<v Speaker 1>innovation on the chips, we'll take advantage of every single

0:04:52.160 --> 0:04:54.960
<v Speaker 1>available GPU and CPU for our customers.

0:04:55.440 --> 0:04:59.600
<v Speaker 2>That's expensive and also, look, investors have been so excited

0:04:59.640 --> 0:05:02.800
<v Speaker 2>around just generative AI full stop, but they want to

0:05:02.839 --> 0:05:05.760
<v Speaker 2>see the revenue. They want to see the profitability chantanute.

0:05:05.760 --> 0:05:08.760
<v Speaker 2>You feel that you're able to give more light, more

0:05:08.800 --> 0:05:11.440
<v Speaker 2>clarity to the market as to when this really becomes

0:05:12.080 --> 0:05:13.279
<v Speaker 2>very revenue generating.

0:05:14.640 --> 0:05:18.240
<v Speaker 1>You're absolutely right, Carolin, that all of the focus thus

0:05:18.320 --> 0:05:20.960
<v Speaker 1>far has been on training, and people have been creating

0:05:21.000 --> 0:05:24.400
<v Speaker 1>all the models, and so all of the investment that's

0:05:24.440 --> 0:05:27.760
<v Speaker 1>actually happening is on the training side of the equation. However,

0:05:28.160 --> 0:05:32.400
<v Speaker 1>I think Adobe has delivered more generative AI software than

0:05:32.480 --> 0:05:36.839
<v Speaker 1>virtually any other large company that had a traditional business.

0:05:36.880 --> 0:05:39.279
<v Speaker 1>So if you think about what we've done with Acrobat

0:05:39.279 --> 0:05:42.800
<v Speaker 1>AI Assistant or Photoshop, and clearly the emphasis has to

0:05:42.839 --> 0:05:46.239
<v Speaker 1>go on inference because that's where people are seeing value,

0:05:46.279 --> 0:05:49.680
<v Speaker 1>that's where it's getting embedded in the workflows. You know,

0:05:49.760 --> 0:05:52.920
<v Speaker 1>we will have three or four different ways in which

0:05:52.960 --> 0:05:55.359
<v Speaker 1>we monetize it, and we have a financial analyst Q

0:05:55.440 --> 0:05:57.640
<v Speaker 1>and A where we will talk about this as well.

0:05:57.640 --> 0:06:01.160
<v Speaker 1>But as a you know, insight into that. Certainly our

0:06:01.200 --> 0:06:04.240
<v Speaker 1>subscription tier pricing will have the ability to lose some

0:06:04.279 --> 0:06:08.000
<v Speaker 1>of this AI. We will have add on services. Firefly

0:06:08.160 --> 0:06:11.880
<v Speaker 1>services is being used by every large enterprise. They want

0:06:11.880 --> 0:06:14.680
<v Speaker 1>to understand how they create a custom model where they

0:06:14.720 --> 0:06:18.200
<v Speaker 1>can have, for example, Bloomberg assets in a custom model

0:06:18.240 --> 0:06:21.039
<v Speaker 1>where only Bloomberg is allowed to generate with that content.

0:06:21.160 --> 0:06:23.520
<v Speaker 1>So I think we have multiple ways in which we

0:06:23.560 --> 0:06:28.520
<v Speaker 1>monetize it today. Already, performance marketers will have a new

0:06:28.520 --> 0:06:31.520
<v Speaker 1>product today which is called gen Studio, so you can

0:06:31.600 --> 0:06:34.719
<v Speaker 1>increase the agility by which you place ads on every

0:06:34.760 --> 0:06:38.000
<v Speaker 1>advertising system. So you know, we're both excited about the

0:06:38.040 --> 0:06:42.280
<v Speaker 1>technology and the value, but we're also confident that customers

0:06:42.279 --> 0:06:44.840
<v Speaker 1>will see the value and enable us to monetize it.

0:06:46.000 --> 0:06:49.640
<v Speaker 3>Chantonnay intense interest in your m and A strategy. Karen

0:06:49.680 --> 0:06:52.520
<v Speaker 3>and I were with the Runway CEO last week. Will

0:06:52.560 --> 0:06:54.200
<v Speaker 3>you look at companies like Runway.

0:06:55.320 --> 0:06:57.800
<v Speaker 1>We're partnering with Runway. You know, I think as it

0:06:57.839 --> 0:07:01.840
<v Speaker 1>relates to models, we believe that Adobe for its core

0:07:02.480 --> 0:07:07.120
<v Speaker 1>domains such as video and imaging and design, and we

0:07:07.160 --> 0:07:10.440
<v Speaker 1>will certainly have our own models, but people will want

0:07:10.520 --> 0:07:13.960
<v Speaker 1>us to support third party models in our applications because

0:07:14.000 --> 0:07:17.480
<v Speaker 1>that is the interface that everybody uses. We've demonstrated that

0:07:17.600 --> 0:07:21.280
<v Speaker 1>for video at IBC with Runway, and certainly we will

0:07:21.280 --> 0:07:23.760
<v Speaker 1>support these custom models. As I said, so if you're

0:07:23.840 --> 0:07:26.720
<v Speaker 1>Coke or Nike or Bank of America and you're using

0:07:26.760 --> 0:07:29.920
<v Speaker 1>it within your context, you have access to that, So

0:07:30.560 --> 0:07:33.320
<v Speaker 1>you know this is going to be such an exciting area.

0:07:33.800 --> 0:07:36.800
<v Speaker 1>Each model, I believe ED will have its own personality.

0:07:37.160 --> 0:07:40.040
<v Speaker 1>So rather than you know, try to create one model

0:07:40.080 --> 0:07:42.840
<v Speaker 1>that's the super set, We're going to support all these

0:07:42.880 --> 0:07:45.960
<v Speaker 1>models because they will do some things differently and better.

0:07:47.240 --> 0:07:50.280
<v Speaker 3>Adobe CEO Shansanuin Ryan is great to finally have you

0:07:50.360 --> 0:07:53.000
<v Speaker 3>here on Bloomberg Technology. Thank you for your time